So, who's getting a 5XXX series card?

[maxfly]

This whole situation is ridiculous to me. We shouldn't have to worry about how the cable is bent, how close to the cards connector or case window the bend radius is, making sure the pins are lined up perfectly... cmon man. This is such a joke. Nvidia has to stop with half measures. Add a 2nd or 3rd or 4th connector and be done with this flaming nonsense. At the least, allow AIBs to design their own upgraded solutions as they see fit.

 

[yoadknux]

All cables melted. Indeed. Mostly garbo ones, we'd agree... cheaper, larger tolerances, less q/a. I'd also be willing to bet some q/a problems on brand name (while i agree with not getting tech from j2c, coincindently, the corsair cable was all but defective in that vid, lol)

The cable bought for a 4090 two years ago is unsafe to use on the new higher power card, yes (it seems, lol), as it doesn't meet the updated requirements (its not h+, even). You'll have to buy another with updated standards that works with the new, higher-wattage, seemingly poorly designed gpu.

....this...... doesn't bother me. I dont think there's anything dubious going on from the cable side. It's them scrambling to get the new standard cables out to market. Of course this benefits them. It's not a chicken or the egg thing for me.

If there's a real revised cable that solves 100% of the issue, then I'm all-in for that and I'm willing to pay. I just don't think such cable exists, and that it's unique to MODDIY. If such cable exists, it means 100% of 4090/5090 users will have to get such a cable, throwing away their old Corsair/CableMod/MODDIY cables. My gut feeling is that the "new ultimate cable" is probably just as bad when you plug it into an H+ card.

 

[EarthDog]

If there's a real revised cable that solves 100% of the issue, then I'm all-in for that and I'm willing to pay.

H++ card + properly made h++ cable = win... or better, no fires.. lol.

The good news is that you get both with 5000 series! If i was a 4090 owner using 3rd party cables/one that didn't come with my card (im using the one that came with my card since before launch day), I'd spend $20 on a new H++, yep. Most 40x0 Supers had them too... though their power envelope doesn't allow them to torch the cables like a 4090 does.




Serious question, to anyone... out of the 3(?) 5090s we've seen go toasty, were any using native h++ (inc. the cable that comes with the card?

This whole situation is ridiculous to me. We shouldn't have to worry about how the cable is bent, how close to the cards connector or case window the bend radius is, making sure the pins are lined up perfectly... cmon man. This is such a joke. Nvidia has to stop with half measures. Add a 2nd or 3rd or 4th connector and be done with this flaming nonsense. At the least, allow AIBs to design their own upgraded solutions as they see fit.

If i were a betting man (I am), we'll see that in the refresh/super cards.. or some official load balancing like previous gens.

I don't know AIBs aren't already allowed to do so but simply haven't.

Edit: reddit mega thread tracking cases (3 - all using non h++ ftr)....

https://www.reddit.com/r/nvidia/comments/1inpox7/rtx_50_series_12vhpwr_megathread

 

[yoadknux]

H++ card + properly made h++ cable = win... or better, no fires.. lol.

The good news is that you get both with 5000 series! If i was a 4090 owner using 3rd party cables/one that didn't come with my card (im using the one that came with my card since before launch day), I'd spend $20 on a new H++, yep. Most 40x0 Supers had them too... though their power envelope doesn't allow them to torch the cables like a 4090 does.




Serious question, to anyone... out of the 3(?) 5090s we've seen go toasty, were any using native h++ (inc. the cable that comes with the card?



If i were a betting man (I am), we'll see that in the refresh/super cards.. or some official load balancing like previous gens.

I don't know AIBs aren't already allowed to do so but simply haven't.

Edit: reddit mega thread tracking cases (3 - all using non h++ ftr)....

https://www.reddit.com/r/nvidia/comments/1inpox7/rtx_50_series_12vhpwr_megathread

There are no H++ cables. Only connectors on the GPU and potentially PSU.

 

[Kenrou]

"The real "User Error" is with Nvidia"

1:08 About the measurements from the last video
3:22 Reproducibility
6:29 Load distribution as the cause?
7:13 Test scenario
9:32 Measurements during operation
11:42 Measurements under load
13:32 Real measurement results
14:42 Test with new cable
15:19 Summary/Conclusion

 

[Iron Bars]

He should have used an Ammeter in-line to solidify his case, especially after cutting wires. I still don't trust clamp meters (from experience). Quite easy to fool them.
Edit: Before you say the connector has sufficient number of mating cycles, remember that the mating cycle number is statistical, not absolute.

 

[Railgun]

So, all this talk about the cables...are the cables not identical? From what I can see, absolutely nothing has changed on the cable side...that is to say the female connector, only the component side, the male side. Pin lengths changed on the component/male side...but I can find zero reference anywhere that the cable socket positions/lengths have actually changed.

 

[dfonda]

H++ card + properly made h++ cable = win... or better, no fires.. lol.

The good news is that you get both with 5000 series! If i was a 4090 owner using 3rd party cables/one that didn't come with my card (im using the one that came with my card since before launch day), I'd spend $20 on a new H++, yep. Most 40x0 Supers had them too... though their power envelope doesn't allow them to torch the cables like a 4090 does.




Serious question, to anyone... out of the 3(?) 5090s we've seen go toasty, were any using native h++ (inc. the cable that comes with the card?



If i were a betting man (I am), we'll see that in the refresh/super cards.. or some official load balancing like previous gens.

I don't know AIBs aren't already allowed to do so but simply haven't.

Edit: reddit mega thread tracking cases (3 - all using non h++ ftr)....

https://www.reddit.com/r/nvidia/comments/1inpox7/rtx_50_series_12vhpwr_megathread

You get a H++ cable with a 5090 ?

 

[Kenrou]

Found the engineer's reddit post he's talking about

"The 8-pin standard asks for 150W at 12V, so 12.5A. Rounding up a bit you might say that it needs 4.5A per pin. With 9-amp connectors, each one is only at half capacity. In a 600W 12VHPWR connector, each pin is being asked for 8.33A already. If you have 8.5A pins, there is functionally no headroom here, and if you have 9A pins, yeah that's not great either. Those 8.5A pins will fail under real-world conditions such as higher ambient temperatures, imperfect surface cleaning, and transient spikes from GPUs. The 9A pins are not much better. (13A pins are probably fine on their own. Margins still aren't as good as the 8-pin, but they also aren't as bad as 9A pins would be.)

I firmly believe that this is where the problem lies. These (not the 13A ones) pins are at the limit, and the margin of error of as little as 1 sixth of an amp before you max out a pin is far too small for consumer hardware. Safety factor here is abysmal. 9Ax12Vx6pins = 648W, and if using 8.5A pins, 612W. The connector itself is good supposedly for up to 660W, so assuming they are allowing a slight overage on each pin, or have slightly better pins than I can find in 5 minutes on the Molex website (they do), you still only have a safety factor of 1.1x."

https://www.reddit.com/r/nvidia/comments/1iogjxp/an_electrical_engineers_take_on_12vhpwr_and

 

[EarthDog]

You get a H++ cable with a 5090 ?

I only have one (Giga Aorus Master), but it came with H++.


...watching Dr8's new vid............

 

[dfonda]

I only have one (Giga Aorus Master), but it came with H++.


...watching Dr8's new vid............

So what exactly is it is it an entire cable going to the PSU? Or is it an adapter like the ones that came with 4000 series.

 

[Railgun]

Found the engineer's reddit post he's talking about

"The 8-pin standard asks for 150W at 12V, so 12.5A. Rounding up a bit you might say that it needs 4.5A per pin. With 9-amp connectors, each one is only at half capacity. In a 600W 12VHPWR connector, each pin is being asked for 8.33A already. If you have 8.5A pins, there is functionally no headroom here, and if you have 9A pins, yeah that's not great either. Those 8.5A pins will fail under real-world conditions such as higher ambient temperatures, imperfect surface cleaning, and transient spikes from GPUs. The 9A pins are not much better. (13A pins are probably fine on their own. Margins still aren't as good as the 8-pin, but they also aren't as bad as 9A pins would be.)

I firmly believe that this is where the problem lies. These (not the 13A ones) pins are at the limit, and the margin of error of as little as 1 sixth of an amp before you max out a pin is far too small for consumer hardware. Safety factor here is abysmal. 9Ax12Vx6pins = 648W, and if using 8.5A pins, 612W. The connector itself is good supposedly for up to 660W, so assuming they are allowing a slight overage on each pin, or have slightly better pins than I can find in 5 minutes on the Molex website (they do), you still only have a safety factor of 1.1x."

https://www.reddit.com/r/nvidia/comments/1iogjxp/an_electrical_engineers_take_on_12vhpwr_and

Kinda what I'd said on Tuesday.

 

[EarthDog]

So, all this talk about the cables...are the cables not identical? From what I can see, absolutely nothing has changed on the cable side...that is to say the female connector, only the component side, the male side. Pin lengths changed on the component/male side...but I can find zero reference anywhere that the cable socket positions/lengths have actually changed.

PCI-SIG. The official document is behind a pay-wall at PCI-SIG.

Or is it an adapter like the ones that came with 4000 series.

An adapter, just like previous gens.


EDIT: Video feels like a rinse and repeat... same poor NV power design (root cause), poor/sub-standard (read 12VHPWR, non H++) cables are bad with a 5090 and 4090.

 

[dfonda]

PCI-SIG. The official document is behind a pay-wall at PCI-SIG.


An adapater, just like previous gens.

Okay thanks. So I'm trying to get this clear the PSU cable itself should then be replaced if it's prior to 2025 which all of my psu cables are. In other words what's the point of having an h++ adapter plugged into an older cable. Or am I missing something?

This would be the reason it's been hard to get Lian li edge psus they were remaking the cables. For the thousand and 1300 Watt psus.

 

[yoadknux]

sigh... i see the confusion struck this topic too even though we discussed it many times already... no such thing as H++ cable



source: https://www.corsair.com/us/en/explo...9kHXHuATugJY_fG6gGybwYzaIfYPLdWfP1o697ObB2ZmJ

 

[Dolk]

Oh you guys can't get to the SIG Spec? I can I wonder if I can post it here. The most you'd get out of the spec is the same information posted else where, just more technical babel.

I'm back and forth on the cable vs plug. I was at Dell in the lab when we first started to test these adapters. There were some issues with power and melty cables for how we were rating them but it seemed like a simple issue to solve. I'm not too sure, someone else on my team was working on that integration. I've seen the same thing that De8aur posted in his first melty cable video where we would ramp the current to see how much we could push through the plugs. I really do not think it is the plug's fault, and I do not think the cable is the issue either. I think they are just the victims to bad power designs on the PCB. Recall that not too long ago someone LN2'd a 5090D and used 1KW at times and there was no melty anything.

 

[EarthDog]

I really do not think it is the plug's fault, and I do not think the cable is the issue either. I think they are just the victims to bad power designs on the PCB. Recall that not too long ago someone LN2'd a 5090D and used 1KW at times and there was no melty anything.

Feels like the root cause is NV's poor design. That coupled with low-quality and/or not in spec (read: 12v-2x6) cables.

sigh... i see the confusion struck this topic too even though we discussed it many times already... no such thing as H++ cable

What is the difference between 12V-2X6 and 12VHPWR?

The latest 12V-2x6 "H++" power connector recently made debut on the NVIDIA GeForce RTX 40 SUPER series GPUs and offers slightly higher power delivery versus the existing 12VHPWR "H+" connectors.

help.moddiy.com

H++ Power Connector = 12V-2x6 Design (675W Power Delivery)
H+ Power Connector = 12VHPWR Design (600W Power Delivery)

but with enhancements in safety and efficiency. Specifically, the power pins have been extended by 0.25mm to enhance connection reliability by enlarging the contact surface area and reducing temperatures. Additionally, the signal pins have been shortened by 1.5mm to ensure users have an easier time fully plugging in the connector to achieve the safest, most reliable connection.

Seems like there is...

 

[yoadknux]

Oh you guys can't get to the SIG Spec? I can I wonder if I can post it here. The most you'd get out of the spec is the same information posted else where, just more technical babel.

I'm back and forth on the cable vs plug. I was at Dell in the lab when we first started to test these adapters. There were some issues with power and melty cables for how we were rating them but it seemed like a simple issue to solve. I'm not too sure, someone else on my team was working on that integration. I've seen the same thing that De8aur posted in his first melty cable video where we would ramp the current to see how much we could push through the plugs. I really do not think it is the plug's fault, and I do not think the cable is the issue either. I think they are just the victims to bad power designs on the PCB. Recall that not too long ago someone LN2'd a 5090D and used 1KW at times and there was no melty anything.

The fact it's so hard to replicate, is in my opinion an indicator that this is an effect that is added over time, not an instantaneous effect.
The connector goes through heat cycles? Gets plugged in and out multiple times? Is bent or routed in a way that stresses the connector? Has trouble making contact because the power pins are too short (H+)? All those things add up and slowly degrade the cable over time until it reaches a critical point and melts.

Post magically merged: Feb 14, 2025

Feels like the root cause is NV's poor design. That coupled with low-quality and/or not in spec (read: 12v-2x6) cables.

What is the difference between 12V-2X6 and 12VHPWR?

The latest 12V-2x6 "H++" power connector recently made debut on the NVIDIA GeForce RTX 40 SUPER series GPUs and offers slightly higher power delivery versus the existing 12VHPWR "H+" connectors.

help.moddiy.com

Seems like there is...

But all those changes are on the GPU connector. The cable is the same. Read the Corsair post

 

[EarthDog]

But all those changes are on the GPU only connector. The cable is the same. Read the Corsair post

MODDIY doesn't make the connector on the GPUs. They are talking about their cables....right?

EDIT: Their connector is different, like Corsair says... the cable, MODDIY at least, supports more power.

EDIT2: When I say cable, I'm talking about the entire thing... connector and wires. My point is that there is a difference between H+ and H++ as you/Corsair pointed out.

EDIT3: I need some caffeine, I see what you're saying now!! Still, **** cables exacerbate the issue.

 

[yoadknux]

MODDIY doesn't make the connector on the GPUs. They are talking about their cables....right?

EDIT: Their connector is different, like Corsair says... the cable, MODDIY at least, supports more power.

EDIT2: When I say cable, I'm talking about the entire thing... connector and wires. My point is that there is a difference between H+ and H++ as you/Corsair pointed out.

EDIT3: I need some caffeine, I see what you're saying now!! Still, **** cables exacerbate the issue.

I think moddiy wrote it vaguely on purpose. You read and think "good, they revised the cable". Then you read Corsair's post and learn that the cable is the same... MODDIY also threw "3.1" into that post too, which is a PSU load specification (atx 3.1). Just complete misinformation and confusion on their part...